Pearu Peterson
Pearu Peterson
Currently, OmniSci UDFs has support only for the following first-class types: ``` bool, int8, int16, int32, int64, float, double ``` and arrays of these types. So, string argument types and...
x-ref: issue #87
It is not immediately clear why this PR affects mkldnn layout behavior. Could you try restoring aten/src/ATen/FunctionalTensorWrapper.cpp to see if mkldnn tests depend on it?
@aartbik , we'll still need to figure out why mkldnn tests fail with this PR. Can you confirm if restoring aten/src/ATen/FunctionalTensorWrapper.cpp will fix the mkldnn tests? Can you reproduce mkldnn...
@aartbik I am OOO til July 28.
Making a redefinition a hard-error can be reasonable when redefinition is done by accident. However, if the redefinition can be supported in Numba such that the redefinition act itself is...
Re: ``` > total_size -= sdiff E OverflowError: Python integer -7812 out of bounds for uint32 ``` I think the resolution here is to make sure that `total_size` (and perhaps...
@QuantFreedom1022 not likely, for instance, cgutils now lives under numba.core. However, it should be possible to fix the feature for newer numba versions.
Just FYI, the corresponding issue in JAX was fixed in https://github.com/jax-ml/jax/pull/27107: see the code around comments " When a is close to zero.." and "For very small a and to...
Adding "mutable arrays" to library level `capabilities` is sub-optimal for libraries that support both mutable and immutable arrays. For example, numpy arrays have `flags.writable` attribute bit that signals if an...